adopt-ai-use-stories git:20261011.dce569cA

Immutable. This exact content never changes and is served at /api/v1/blob/2daaed551dd3f634.

---
name: adopt-ai-use-stories
description: Prepares a consent note, an interview guide and three to five short anonymous accounts of real AI use at work, each drafted from the teller's own words, approved by the teller and marked as one case. Use for "run adopt-ai-use-stories", "AI use stories", "collect stories of AI use", "real examples of AI at work", "AI case stories from our teams", "interview guide for AI use", "consent note for a use story", "examples that are not demos", part of the Claude Guide for AI Adoption Pack by Polar Bear.
---

# AI Use Stories

## When To Use
The numbers are thin and the examples in circulation are vendor demos. Use this to collect a few accounts of what colleagues really did with AI on real work, told with permission and without names. It answers: what does a real use look like here, including the one that did not work?

## When Not To Use
Not to show how widespread anything is: a story never produces a rate, so use AI Adoption Evidence Review for counts. Not to list what AI could do for a role, which is AI Use Case Map. And not when a small team makes any teller recognisable and broadening the details would empty the story: collect nothing and say why.

## Inputs
- The teams or roles in scope, the period you are looking back over, where the stories will be shared, and who will collect them.
- After the conversations: the collector's notes or transcript for each teller, with their agreement to draft from them.
If you have none of this, I start from the scope in one sentence, write the consent note and the interview guide, and mark the output as a first draft with no stories in it. I never write a story without a teller's words.

## Approach
Collecting stories of change, from the Most Significant Change technique described in Davies and Dart's public guide: define the domains of change first, collect with an open question, record who collected each story and when, verify it with the teller, and feed back how the stories were used. This guide adds its own safeguards and leaves out any step that picks a "best" story or chooses tellers by results. A person holds each conversation; I prepare the note and the guide, then draft from what the teller said. The failure this prevents is the success story lifted from a keen colleague's chat, polished by someone else, shared with enough detail that the whole floor knows who it is, and quoted the next month as proof that everyone benefits.

## Workflow
1. Ask at most three questions: which teams or roles and which period, where will the stories be shared, and who collects them?
2. Set two to four domains of change to look for: a task done differently, a check added, something stopped, something that did not work.
3. Write one invitation in the same wording for everyone in scope. Nobody is approached because of how much or how well they use AI, and no usage data is used to find tellers.
4. Write the consent note: what will be written, where it will be shared, that the teller reads and approves the final text, that they can withdraw before it is shared, that no name and no identifying detail appears, and that saying no has no consequence.
5. Write the interview guide: one open question per domain, then what did you try, what did you check, what changed, what did not work. The collector asks and listens; the guide is not a form to complete.
6. From the notes you paste, draft each story in the teller's words under the five headings, within the length you set. Where a detail could identify the teller in a small team, broaden it or drop the story. Mark every gap instead of filling it.
7. Send each draft back through the collector for the teller's approval, keep at least one story about something that did not work, end each with "One case, not a rate.", and plan the feedback to tellers.

## Output Format
```markdown
# AI Use Stories
**Scope:** [teams or roles invited] · **Period:** [dates] · **Collected by:** [role] · **Domains of change:** [two to four]
## Consent note
We are collecting short accounts of how people have used AI in their work, including what did not work. If you take part: we write up [what], and share it [where, with whom]. You read the final text and nothing is shared until you approve it. You can withdraw before [date it is shared]. No name and no identifying detail appears. Saying no has no consequence. Questions go to [name, role].
## Interview guide
1. Looking back over [period], what changed most in how you do [task]? [One open question like this for each further domain.]
2. What did you try? What did you check? What changed? What did not work?
## Stories
### Story [n]: [the task, in general terms]
- **Domain:** [domain of change]. Collected by [role] on [date]. Approved by the teller on [date].
- **The task:** [what the work was]
- **What was tried:** [in the teller's words]
- **What was checked:** [how the output was verified]
- **What changed:** [in the work]
- **What did not work:** [and what the teller did instead]

One case, not a rate.
## Feedback to tellers
- Story [n]: shared in [place], used for [use], teller told on [date] by [role].
## Decision
Each teller decides whether their own story is shared, and can withdraw before [date]. [Named owner] decides by [date] which approved stories go where, and reports back to tellers by [date].
```

## Done When
- The invitation went to everyone in scope in the same wording, and the consent note covers all six points.
- There are three to five stories, each with the five headings, a collector, a date and the teller's approval.
- At least one story is about something that did not work, every story ends with "One case, not a rate.", and no story could identify its teller.

## Quality Bar
- I never invent or embellish a story, a quote or a result. A draft holds only what the teller said, and a missing part is a bracketed gap.
- No figure for time saved appears unless the teller gave it, and then it is theirs, about one task, and labelled as such.
- The stories are not ranked, no "best" story is voted, tellers are never compared, and stories are never counted to make a percentage.
- A withdrawn story is removed everywhere, without a reason being asked.
- Voluntary, consented, anonymous, never chosen by ranking people, and each story is one case, not a rate.

## Next
Run adopt-executive-update (AI Adoption Executive Update) to put the evidence and one or two stories on one page.

## About the makers

This pack is made by Polar Bear, a consultancy built by ex-McKinsey founders with a dream to make AI work for People, not instead of them. We help our clients build people systems and AI-first ways of working, and we run our own company on Claude. If your team has outgrown the self-serve version, message Pauline (linkedin.com/in/paulinebertry).